A wavelet-based feature vector model for DNA clustering.

Abstract

DNA data are important in the bioinformatic domain. To extract useful information from the enormous collection of DNA sequences, DNA clustering is often adopted to efficiently deal with DNA data. The alignment-free method is a very popular way of creating feature vectors from DNA sequences, which are then used to compare DNA similarities. This paper proposes a wavelet-based feature vector (WFV) model, which is also an alignment-free method. From the perspective of signal processing, a DNA sequence is a sequence of digital signals. However, most traditional alignment-free models only extract features in the time domain. The WFV model uses discrete wavelet transform to adaptively yield feature vectors with a fixed dimension based on the features in both the time and frequency domains. The level of wavelet transform is adjusted according to the length of the DNA sequence rather than a fixed manually set value. The WFV model prefers a 32-dimension feature vector, which greatly promotes system performance. We compared the WFV model with the other five alignment-free models, i.e., k-tuple, DMK, TSM, AMI, and CV, on several large-scale DNA datasets on the DNA clustering application by means of the K-means algorithm. The experimental results showed that the WFV model outperformed the other models in terms of both the clustering results and the running time.